Target detection method based on non-convex motion assistance

Abstract

The invention discloses a target detection method based on non-convex motion assistance. The method comprises the following steps: step 1, inputting to-be-processed video data into a constructed low-rank sparse decomposition model; 2, solving the constructed low-rank sparse decomposition model by adopting an alternating direction multiplier method to obtain a moving target in the input video data;according to the method, a non-convex gamma norm is adopted to replace a rank function in a traditional low-rank sparse decomposition model to approximately represent a low-rank part of a video background; meanwhile, by considering that the background still has sparsity in the transform domain, the l1 norm is adopted to carry out sparse approximation on the background of the transform domain, andthe l1 norm representing the sparse prior of the moving target is still adopted as the foreground. Moreover, a motion auxiliary information matrix is introduced into the model, so that the motion auxiliary information matrix is fused into foreground motion information, and the motion target detection of the video is better realized.

Description

technical field [0001] The invention belongs to the field of computer vision, in particular to an object detection method based on non-convex motion assistance. Background technique [0002] Moving target detection is one of the most important and challenging tasks in the field of computer vision, and its performance directly affects subsequent processing such as target tracking and target recognition. Low Rank and Sparse Decomposition (LRSD), also known as Robust Principal Components Analysis (RPCA), has been widely concerned in the fields of computer vision, especially in the detection of moving objects, showing great potential. In order to accurately detect moving objects, many problems are faced in the actual scene, such as lighting changes, camouflaged objects, shaking leaves, water surface ripples, fountains, etc. The judgment of the target is prone to errors, and the extracted background noise is large, which is difficult for foreground detection. affect the accuracy...

AI Technical Summary

Problems solved by technology

[0022] Object of the present invention: The present invention aims at the problem of low detection accuracy when traditional low-rank sparse decomposition is used for video moving target detection, and proposes a target detection algorithm based on non-convex motion-assisted low-rank sparse decomposition to realize the rank function Approximate expression depiction is more accurate, achieving better moving target detection

Images